Retrospective On Hiyajo Maho's Conversation About AI - (VN: Steins;Gate 0)

No Story Spoilers are in this article!!

Steins;Gate 0 anime and Visual Novel explore expanded perspectives and ideas over the original Steins;Gate anime and Visual Novel. The main plot device in this sequel is an artificial intelligence system called Amadeus.

I played Steins;Gate 0 last year, and a particular conversation in it felt very relevant to the times we're living right now. Maho's lecture on artificial intelligence. Below, I'll quote parts of the conversation and my take on it with what we know today in 2026.

Maho says that the research on AI (Artifical Intelligence) began at the same time as computers first appeared. Scientists who were amazed by computers concluded if they created something that can think like humans but as fast as a computer, they'll be able to solve many more problems that stumbled researchers.

According to Maho, even though a true intelligent AI hasn't been made yet, researches have already managed to solve many great problems due to the advancement they made on the way there... She even named examples like OCR technology, and Voice Recognition.

The events of Steins;Gate 0 take place in the year 2011, so we should expect a level of technology comparable to 15 years ago. For Maho, I believe something like ChatGPT will be an exciting milestone to that goal...

Still, no one succeeded in creating a True Artificial Intelligence. That would be true even today. Most advanced AI technologies we have in 2026 are the LLMs, which are impressive pattern recognition technologies that succeed at mimicking the capability to reason, without actually having it.

Nowadays, AI research corporations dream of reaching AGI (Artificial General Intelligence), but not only it's poorly defined threshold, (machine's surpassing humanity in all cognitive tasks,) but even if the current LLM architecture is able to reach that level, or it may be too expensive.

Personally, I don't think the Tansformer technology which LLMs are built on can ever get us there.

Okabe asks Maho why no one succeeded in making AI that emulates the human brain, and Maho answers "There's a lot we don't know." One thing she mentioned is "How can a bio-computer made of just 100 billion neurons have such power?" This statement made me think... If you think of LLM parameters as neurons like the human brain, then AI scientists just need to figure out how these neurons interact with each other.

https://inleo.io/threads/view/ahmadmanga/re-leothreads-hbe5518l

https://inleo.io/threads/view/askrafiki/re-ahmadmanga-1770904051

Amadeus, the fictional AI system of Steins;Gate 0 was made by copying the human brain, so even though it works like a human brain would, its creators still don't understand how it works. One of Maho's goals in-universe is to use Amadeus to have a better understanding of the human brain. After all, she's not an AI researcher, she's a neuroscientist!

The "Box on the Table" Experiment

To drive how much a computer is different from a human, Maho suggested an experiment. Moeka played the role of an Andriod, and Okabe had to order her to open a box on the table. He had to be specific, and only words are accepted, so you can't point to the box and say "open that box." An android needs precise instructions.

Okabe tried multiple ways of ordering the android to open the box, but every time, Moeka as the android kept saying she can't. She couldn't know how. She didn't know which box. The box had "this is not a box" written on it. The box isn't pure white as Okabe had said... There are too many parameters that humans intuitively know that an Android needs to be told to work with.

Things we don't even pay mind to, like which day of the week is it? Is it raining right now? What's the air pressure in the room? What's the lighting situation? Things like that can have an effect on how an android opens a box on the table, but humans intuitively know which "parameters" to ignore depending on what their goal is.

https://inleo.io/threads/view/ahmadmanga/re-leothreads-dwwijfoe

You see that even with LLMs today. They have billions of parameters that they check to generate each word as you request them. The biggest breakthrough for LLMs so far are the ways we discovered to reduce the number of using parameters that are useless to the task at hand. That's why you see models with 4 Billion parameters in 2025 beating models with 30 Billion parameters that used older architectures from 2024 and before.

Another example I love is the Mixture of Experts model that allows LLMs to use fractions of their overall knowledge at runtime, reducing the amount of compute needed while maintaining quality: It's like telling AI, think only of the variables that have an effect on this situation we have.

It's fascinating stuff!

What do you think?

I really love talking about AI, even though I see a lot of dangers it brings with it. (And honestly, humans are probably not ready for it yet.) AI is here to stay, so rather than being worried about the dystopia we might devolve to, (that we arguable are already living in,) I rather look at the positives of the technology like Maho did in Steins;Gate 0.

Above all... AI, and many of the technologies humanity was able to create proves how much humanity is able to accomplish with the little brains God has given us. It's just another wonder of life, despite everything bad we were responsible for...

Posted Using INLEO